Any LLM Voice

Prompts to Evaluate a Text-to-Speech Model Properly

Fish Audio's Drama 3 landed this week at 93.2% on a pronunciation robustness benchmark — which tells you about pronunciation robustness and nothing else. Benchmark scores do not predict whether a voice works for your script, your languages or your budget. These five prompts build the test set a leaderboard cannot give you.

⌨️ 5 prompts 🕐 Updated Oct 15, 2026

💡 How to use these prompts: Replace everything in [BRACKETS] with your specific details before sending. Click Copy to copy any prompt to your clipboard instantly.

Build the hard-word test set

Creating a repeatable TTS test set

My content is about [DOMAIN] for [AUDIENCE]. Generate 40 test sentences specifically designed to break a text to speech model: proper nouns, acronyms read as letters versus words, numbers in several formats, units, homographs where stress changes meaning, and two sentences where a comma changes the meaning. Keep each under 25 words. Output as a numbered list I can paste into any TTS tool.

Test code-switching and names

Multilingual and regional pronunciation testing

My listeners are in [REGIONS] and my scripts mix [LANGUAGE A] with [LANGUAGE B] terms. Write 20 sentences that switch languages mid sentence the way my audience actually speaks, plus 10 containing names from [REGIONS] that are commonly mispronounced. Mark where each sentence switches so I can check whether the model handled the boundary.

Score the output like a listener

Comparing TTS output side by side

I have generated the same 10 sentences in three TTS models. Here are my notes on each: [PASTE]. Build a scoring table across pronunciation accuracy, pacing, emotional range, artefacts, and consistency across takes. Weight them for [USE CASE] and tell me which model wins and which single weakness would most annoy my listeners.

Read the licence before the benchmark

Open-weight and community licence review

Here is a TTS model licence: [PASTE]. Answer in plain language: can I use this commercially, at what revenue threshold does that change, may I resell or expose it as a service, what happens to audio I already generated if I cross the threshold, and what is banned outright. Quote the clause for each answer. Flag anything ambiguous enough to need a lawyer.

Work out the real cost per hour

Budgeting voice generation at volume

I need [N] minutes of finished audio per month in [LANGUAGES]. One vendor charges per character and another per second. Pricing: [PASTE]. Estimate characters per minute of speech, show the monthly cost both ways, and add a realistic retake rate for the sentences that need regenerating. Show the arithmetic so I can change my assumptions.